this is both on a compute node in a cluster allocated by SLURM (but no virtualization/cpusets etc) and on machines directly ssh’ed into. Screenshot example:
CentOS 5.10, 16 core (2 socket) machine, direct ssh login, trivial omp hello world example.
Also verified on a system where I am just a regular user and didn’t install anything (6 core * 2 socket = 12 core node reports: cores per socket = 4, total cores = 3).
Doesn’t appear to be anything to do with execstack message on libjpgprof64.so.
Should note this is for data collected by pgcollect. Using -Mprof gave me the right core/socket count. (Reporting on behalf of someone, so don’t have 100% of the info immediately to hand.)